BUG #19116: Lost concurrent updates using ctid - as per doc guidance

From: PG Bug reporting form <noreply(at)postgresql(dot)org>
To: pgsql-bugs(at)lists(dot)postgresql(dot)org
Cc: bernice(dot)southey(at)gmail(dot)com
Subject: BUG #19116: Lost concurrent updates using ctid - as per doc guidance
Date: 2025-11-17 14:31:11
Message-ID: 19116-079f845bd20241db@postgresql.org
Views: Whole Thread | Raw Message | Download mbox | Resend email
Thread:
Lists: pgsql-bugs

The following bug has been logged on the website:

Bug reference: 19116
Logged by: Bernice Southey
Email address: bernice(dot)southey(at)gmail(dot)com
PostgreSQL version: 18.1
Operating system: Linux Mint 22.2
Description:

Here's a contrived exampled to reproduce. (The advisory locks are to force
concurrency.)

--session 1
CREATE TABLE t(p BOOL, q BOOL);
INSERT INTO t DEFAULT VALUES;
SELECT pg_advisory_lock(1);

--session 2
SELECT pg_advisory_lock_shared(1);
SELECT pg_advisory_unlock_shared(1);
WITH lock_t AS (SELECT ctid FROM t FOR UPDATE)
UPDATE t SET p = TRUE FROM lock_t l WHERE t.ctid = l.ctid
RETURNING p, q, t.ctid;

--session 3
SELECT pg_advisory_lock_shared(1);
SELECT pg_advisory_unlock_shared(1);
WITH lock_t AS (SELECT ctid FROM t FOR UPDATE)
UPDATE t SET q = TRUE FROM lock_t l WHERE t.ctid = l.ctid
RETURNING p, q, t.ctid;

--session 1
SELECT pg_advisory_unlock(1);

Only one of the updates succeeds. If using a standard column, instead of
ctid, then both updates succeed.

I was using a similar approach to the one advised in the UPDATE doc final
example[1]. This guidance was added in PostgreSQL 17 (see [2] for the
discussion). This wouldn't matter as used in the example, as subsequent
updates will find the lost updates. But it does matter, if this approach is
used to update a sync record (e.g. a balance).

I first thought this was the same issue as here [3] and wrote a post asking
for confirmation in the general mailing list. But I've now established that
patch was released in 17.7. I'm logging a bug because I think the doc
guidance might be dangerous.
I can reproduce this in:
PostgreSQL 18.1 (Ubuntu 18.1-1.pgdg24.04+2) on x86_64-pc-linux-gnu, compiled
by gcc (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0, 64-bit
PostgreSQL 17.6 (Debian 17.6-2.pgdg13+1) on x86_64-pc-linux-gnu, compiled by
gcc (Debian 14.2.0-19) 14.2.0, 64-bit

1 [https://www.postgresql.org/docs/current/sql-update.html]
2
[https://www.postgresql.org/message-id/flat/CADkLM%3DcaNEQsUwPWnfi2jR4ix99E0EJM_3jtcE-YjnEQC7Rssw%40mail.gmail.com]
3
[https://www.postgresql.org/message-id/flat/4a6268ff-3340-453a-9bf5-c98d51a6f729%40app.fastmail.com]

Browse pgsql-bugs by date

  From Date Subject
Next Message Richard Guo 2025-11-18 08:27:40 Re: GROUP BY ROLLUP queries on views trigger full table scans (index usage not optimized)
Previous Message PG Bug reporting form 2025-11-17 12:56:35 BUG #19115: the package got zero size